Legged Locomotion Seminar 16-899E

Course Name: Seminar on Legged Locomotion
Course Number 16-899 E
Units: 12
Day: Monday, First meeting Jan. 13
Time: 5-6pm
Location: NSH4201
Instructors: Christopher Atkeson, Jessica Hodgins

The goal of this seminar is to understand and move forward the state of the art in robot legged locomotion. The NSF has funded a human-sized bipedal robot for CMU, and we need to be ready to control it when it arrives. We also have part of the Rhex legged locomotion project here at CMU (/www.ri.cmu.edu/projects/project_485.html). Participants will read and present key papers, explore research issues in simulation, and ideally test ideas on actual robots. We are also interested in insights into human locomotion and how to program graphical characters.

Other courses, reading groups

Swiss ANWM_Seminar
LDG group at ATR
MIT Leg Lab


All the topics we could think of.

Grand Challenges

See AIBO Biped Locomotion

Assignment 1

Assignment 1: Find a control scheme that minimizes the cost for walking for 20 seconds for the "compass" or "stiff-legged" biped. You can use the software below, or you can write your own. The cost function is defined in the software below (see compass1.c).

All the software in .tar.gz form

compass.sd description of robot for SDFAST
compass_sar.c SDFAST output
compass_dyn.c SDFAST output
sdlib.c standard SDFAST support routines
compass1.c my interface to SDFAST

demo-policy1.c demo program to run a policy (controller)
demo-pdw.c demo program to run passive dynamic walking (sideways gravitational force simulates tilted walkway)
compass1.h .h file
compass1-policy.c a simple policy (controller)
compass1-policy.h .h file

compass1-graphics.c X11 graphics.
no-graphics.c compile with no-graphics.c to run on non-X11 systems.
useful.c some useful routines.

Assignment 2

Design an "intuitive" (Raibert style) controller for the compass biped.

Assignment 3

Demonstrate the compass biped passive dynamic walking generated by the demo-pdw.c demo program is stable (ideally by estimating eigenvalues (Floquet multipliers))

Bonus: Find the maximally stable passive dynamic walking pattern for the compass biped (define stable).

Bonus 2: Find the maximally robust passive dynamic walking pattern for the compass biped (define robust).